VISTA3D — Worker pipeline (v2)

Same VISTA3D Phase 1 (auto branch) as v1 but with all WASM + WebGPU compute moved to a dedicated Web Worker — the UI thread stays fully responsive (slice scrubbing, scrolling, dropdown work) during the ~90 s sliding window. · Optimizations: canonical CT + accumulator + scratch buffers all pinned to the WASM heap, zero JS↔heap copies per patch; pre-built gaussian; per-patch progress.

✓ Conformance (Node M1, 2026-04-18): per-patch rel_rms 1.5e-6 vs PyTorch; full 213³ sliding rel_rms 4.4e-7, binary agree 100.0000% vs MONAI SlidingWindowInferer.

⚠ Download: ~786 MB of model weights + 28 MB canonical CT will be fetched on Init. Streaming progress shown below.

Class:

Log

Axial slice viewer

Slice w: 0

CT canonical input

Predicted mask (sigmoid > 0.5)

Overlay